Suggested by ZeroKelvin
Created as described in the tutorial: https://civarchive.com/models/52697/tutorial-konyconi-style-lora
Perpetual coffee request: Keep me motivated and caffeinated:
Description
FAQ
Comments (9)
Have you thought about making your loras smaller, like in this tutorial? https://civitai.com/models/22530/guide-make-your-own-loras-easy-and-free
"The dim/alpha mean the size and scaling of your Lora, and they are controversial: For months everyone taught each other that 128/128 was the best, and this is because of experiments wherein it resulted in the best details. However these experiments were flawed, as it was not known at the time that lowering the dim and alpha requires you to raise the learning rate to produce the same level of detail. This is unfortunate as these Lora files are 144 MB which is completely overkill. I personally use 16/8 which works great for characters and is only 18 MB."
I've experimented with lower dim/alpha but it led to weaker results. But I'll definetly check advices in the tutorial.
Training chars is not the same as training styles and as always there is no a perfect settings for training since depends on color, hue, details, style, etc. Of the inputs.
Kony tried 16/1 on constructionyardAI (I think he should have tied 8 alpha there 🙂)
TYSM!
how about Javanese or balinese one?
you could search for statue, shadow puppets, festival or pop culture art for reference.
check out "samurai punk" for Japanese.
@Taco360 It's not japanese, but jaVanese -- I've made the javapunk to respond this suggestion.
@konyconi oops misread Javanese as Japanese lol.
Wow..nice thing you have here , with Bali and Java Punk m you`re indeed starting new trend . I`ll posting the experiment result combining those three . I`ll bet that will be awesome . Thanks for releasing those pack man
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.